IOSR Journal Of Environmental Science, Toxicology And Food Technology (IOSR-JESTFT) e-ISSN: 2319-2402,p- ISSN: 2319-2399. Volume 3, Issue 5 (Mar. - Apr. 2013), PP 01-08 www.Iosrjournals.Org www.iosrjournals.org 0 | Page Forecasting Criteria Air Pollutants Using Data Driven Approaches; An Indian Case Study Tikhe Shruti S. 1 , Dr. Mrs. Khare K. C. 2 , Dr. Londhe S. N. 3 1 (Department of Civil Engineering, Sinhgad College of Engineering, Pune, Maharashtra, India) 2 (Department of Civil Engineering, Sinhgad College of Engineering, Pune, Maharashtra, India) 3 (Department of Civil Engineering, Vishwakarma Institute of Information Technology, Pune, Maharashtra, India) Abstract : Forecasting air pollutant trends especially in metropolitan cities of India, has become a vital issue as air pollution has immediate and severe impacts on human health. Criteria pollutants like Oxides of Sulphur (SOx), Oxides of Nitrogen (NOx) and Respirable Suspended Particulate Matter (RSPM) have either reached or exceeded the acceptable limits specified by Central Pollution Control Board of India for most of the cities like Pune. In the present work two soft computing approaches namely Artificial Neural Networks (ANN) and Genetic Programming (GP) are used to predict the air quality parameters (SOx, NOx, RSPM) a few time steps in advance for Pune City. Six models have been developed based on daily average data values of pollutant concentrations spanning over seven years. ANN, one of the proven tools in estimation and prediction of air quality has been used and the results of the models are compared with GP which is rarely used tool in the field of air quality modelling and forecasting. The performance of the models has been compared using r, RMSE and d. Considering the complexity of the air pollution phenomenon, it was found that GP models are robust and could work well as compared to ANN. Keywords - Air Quality, ANN, Criteria Pollutants, GP I. Introduction Air pollution is a complex issue, fuelled by multiple sources ranging from vehicular exhaust, industrial emissions, emissions from fossil fuels, construction activities to domestic activities. Air pollution may cause pernicious effects on human health, especially in areas with high population density. Forecasting air quality is one of the most sought after topic of research today for urban air pollution studies and specifically for prediction of pollution episodes i.e. high pollutant concentrations causing adverse health effects [1]. Air quality models play a vital role in all aspects of air pollution control and air quality planning, where prediction is a major component [2]. Air quality forecasts provide the public with air quality information which allows people to take precautionary measures to avoid or limit their exposure to unhealthy levels of air pollution. Hence it is quite essential to predict criteria pollutants. Urban air pollution involves physical and chemical process ranging over a wide scale of time and space. In order to model the urban systems, extensive data such as emissions from various sources (stationary and mobile), influence of buildings and other obstacles, meteorology of the area, information about turbulence profile, heat flux , previous values of the pollutants etc. is required. It is practically very difficult to collect the above-mentioned data (except pollutant concentrations), hence temporal models are handy in such situations. They can be used easily for forecasting purpose because historical sequence of the pollutant concentrations is readily available from pollution control authorities of the country. Air pollution is a time dependent phenomenon which further justifies the use of time series approach for forecasting criteria air pollutants. Several techniques are available to predict future pollutant concentrations, including fixed box methods, linear regression methods, computational fluid dynamics (CFD) simulation, artificial intelligence etc. Conventional technique like numerical method require detailed source information and consume a lot of time and effort to forecast and also found to be weak particularly when used to model nonlinear systems [3]. This leaves a scope for another approach like data driven techniques which are found to be suitable to model nonlinear systems. Artificial Neural Networks (ANN) are already been regarded as a cost effective method to achieve the prediction of air pollutants in time series and have become popular since last decade [4]. The literature reveals that Genetic Programming (GP) a relatively new approach has been applied successfully to solve complex Civil Engineering problems such as structural optimisation, soil classification, prediction of scour depth of circular piles, algal bloom prediction and also for prediction of climate change. Better predictive capabilities of GP have also been reported, especially for the peak values for wave forecasting [5]. Air pollution happens to be the